Device fault diagnosis method and device based on active learning, and medium
By introducing active learning methods in device fault diagnosis, screening and updating training samples, and iteratively updating models, the problem of traditional methods relying on professional knowledge and deep learning models with low accuracy under limited data is solved, and higher diagnostic accuracy and model generalization capabilities are achieved.
Patent Information
- Application Number
- CN202510057394.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-14
- Publication Date
- 2025-05-13
AI Technical Summary
Traditional equipment fault diagnosis methods rely on deep expertise and rich experience, and are inconvenient when dealing with complex signals and noises. Deep learning models are difficult to train high-quality fault diagnosis models under limited data, resulting in low accuracy.
Using the equipment fault diagnosis method based on active learning, the uncertainty measurement value of the second historical fault vibration signal is determined by judging the performance of the trained one-dimensional fault diagnosis model, the training samples are screened and updated, and the model iteratively is updated to improve the diagnostic accuracy.
Through the active learning process, the training samples are effectively expanded, the accuracy of model training is improved, the generalization ability and robustness of the model are enhanced, and the comprehensiveness and accuracy of equipment fault diagnosis are improved.
Smart Images
Figure CN119989145A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of equipment fault diagnosis, and in particular to an equipment fault diagnosis method, equipment and medium based on active learning. Background Art
[0002] The core position of equipment fault diagnosis in the field of asset and equipment management is becoming increasingly prominent. Its importance is not only reflected in maintaining the stable operation of the production line, but also directly related to the company's operational efficiency, cost control and safe production. With the rapid development of industrial automation and the continuous innovation of equipment technology, modern industrial systems have become more and more complex, integrating a large number of sensors, actuators and control systems, which makes the interaction between devices more frequent, but also increases the risk of potential failures. Therefore, it has become an urgent need to respond quickly to abnormal conditions of equipment.
[0003] Traditional equipment fault diagnosis methods mainly rely on the fine processing and analysis of equipment signals, such as time domain analysis, frequency domain analysis, wavelet transform, etc. These methods can reveal hidden features in the signal and thus infer the health status of the equipment. However, these methods often require deep professional knowledge and rich experience accumulation, and have limited ability to identify signal noise and complex faults. In addition, as the complexity of the equipment increases, the coupling relationship between signals becomes more complex, and traditional methods are not convenient in dealing with such problems. In order to overcome the limitations of traditional methods, machine learning and deep learning technologies have been introduced into the field of equipment fault diagnosis. These technologies can automatically learn complex patterns from a large amount of data and perform tasks such as classification and prediction based on them. In equipment fault diagnosis, deep learning models can capture high-level features in signals, which often reflect the true status of the equipment better than manually designed features. However, the training of deep learning models requires a large amount of labeled data as support. However, equipment failures are often sporadic, and the types of failures are diverse, making it difficult to fully cover them. Accurate labeling of fault data requires professional knowledge and experience, which increases the cost and difficulty of data collection. Therefore, it is difficult to train a high-quality fault diagnosis model using deep learning methods with limited data, resulting in low accuracy in equipment fault diagnosis. Summary of the invention
[0004] In order to solve the above technical problems, one or more embodiments of this specification provide a device fault diagnosis method, device and medium based on active learning.
[0005] One or more embodiments of this specification adopt the following technical solutions:
[0006] One or more embodiments of this specification provide a device fault diagnosis method based on active learning, the method comprising:
[0007] Determine whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance; wherein the first historical fault vibration signal is a fault vibration signal marked with a fault label on a single part of the equipment to be detected;
[0008] If not, then determining the uncertainty measurement value of each of the second historical fault vibration signals according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model; wherein the second historical fault diagnosis signal is a fault vibration signal on a single part of the equipment to be detected that is not marked with a fault label;
[0009] screening the second historical fault diagnosis signals according to the uncertainty measurement value of each of the second historical fault diagnosis signals, so as to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signals;
[0010] Based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain a one-dimensional fault diagnosis model for each part, so as to identify and diagnose the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part.
[0011] Optionally, in one or more embodiments of the present specification, determining the uncertainty measurement value of each of the second historical fault vibration signals according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model specifically includes:
[0012] Determining the probability and fault diagnosis result of each fault category corresponding to the second historical fault diagnosis signal according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model;
[0013] Determine the maximum probability of the fault category of the trained one-dimensional fault diagnosis model at the location according to the probability of each fault category;
[0014] Summing the difference between the preset data and the maximum probability corresponding to each part to obtain an initial first uncertainty measurement value, and normalizing the initial first uncertainty measurement value to obtain a first uncertainty measurement value;
[0015] Determining the model prediction probability corresponding to the trained one-dimensional fault diagnosis model based on the fault diagnosis result, performing standardized summation on the variances of the model prediction probabilities corresponding to the various parts to obtain an initial second uncertainty measurement value, and performing normalization processing on the initial second uncertainty measurement value to obtain a second uncertainty measurement value;
[0016] The uncertainty metric value of each of the second historical fault vibration signals is determined by combining the first uncertainty metric value and the second uncertainty metric value.
[0017] Optionally, in one or more embodiments of the present specification, before determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes:
[0018] Inputting the first historical fault diagnosis signal into the one-dimensional convolution layer of the initial one-dimensional fault diagnosis model, traversing the first historical fault vibration signal according to the fixed stride of the one-dimensional convolution layer and the preset convolution kernel, and obtaining a convolution output vector corresponding to the first historical fault diagnosis signal;
[0019] Determine the one-dimensional output vector corresponding to the convolution output vector based on the fully connected layer of the initial one-dimensional fault diagnosis model, and input the one-dimensional output vector into the softmax layer of the initial one-dimensional fault diagnosis model to determine the probability of each fault category corresponding to the one-dimensional output vector;
[0020] The output result of the initial one-dimensional fault diagnosis model is determined based on the probability of each of the fault categories, and it is determined whether to perform iterative training on the initial one-dimensional fault diagnosis model according to the output result to obtain a trained one-dimensional fault diagnosis model.
[0021] Optionally, in one or more embodiments of the present specification, after determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes:
[0022] If it is determined that the trained one-dimensional fault diagnosis model meets the preset model performance, the trained one-dimensional fault diagnosis model is used as the one-dimensional fault diagnosis model of the corresponding part;
[0023] The fault diagnosis signal of the corresponding part of the equipment to be detected is input into the one-dimensional fault diagnosis model to obtain the fault diagnosis result of the fault diagnosis signal.
[0024] Optionally, in one or more embodiments of the present specification, the second historical fault diagnosis signals are screened according to the uncertainty measurement value of each second historical fault diagnosis signal to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signal, specifically including:
[0025] sorting the second historical fault diagnosis signals based on the uncertainty measure value of each of the second historical fault diagnosis signals;
[0026] sequentially obtaining a preset proportion of second historical fault diagnosis signals based on the sorting order;
[0027] The second historical fault diagnosis signal with the preset ratio is sent to a preset expert terminal for labeling to obtain a second historical fault diagnosis signal labeled with a fault label, and the first historical fault diagnosis signal is updated with the second historical fault diagnosis signal with the preset ratio.
[0028] Optionally, in one or more embodiments of the present specification, based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain a one-dimensional fault diagnosis model for each part, specifically including:
[0029] The trained one-dimensional fault diagnosis model is trained based on the updated first historical fault diagnosis signal, and the second historical fault diagnosis signal corresponding to the fault vibration signal not marked with a fault label is updated according to the filtered second historical fault diagnosis information;
[0030] The training set is iteratively updated, so as to iteratively update the trained one-dimensional fault diagnosis model based on the updated training set, so as to obtain a one-dimensional fault diagnosis model of each part that meets the requirements.
[0031] Optionally, in one or more embodiments of the present specification, before identifying and diagnosing the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part, the method further includes:
[0032] Based on the preset collection devices of various parts of the equipment to be detected, the fault vibration signals of various parts are obtained by collecting data based on the time sequence;
[0033] The fault vibration signal is segmented based on the input size corresponding to the one-dimensional fault diagnosis model of each part to obtain the time series data of the vibration signal to be input.
[0034] Optionally, in one or more embodiments of the present specification, identifying and diagnosing the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part specifically includes:
[0035] Inputting the time series data of the vibration signal to be input into the one-dimensional fault diagnosis model of the corresponding part, so as to determine the fault category of the fault vibration signal according to the probability of each fault category output by the one-dimensional fault diagnosis model of the corresponding part;
[0036] The fault categories obtained by the one-dimensional fault diagnosis model of each part are voted, and the fault category with the most votes is selected as the final fault diagnosis result.
[0037] One or more embodiments of this specification provide a device fault diagnosis device based on active learning, the device comprising:
[0038] at least one processor; and,
[0039] a memory communicatively connected to the at least one processor; wherein,
[0040] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned methods.
[0041] One or more embodiments of the present specification provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute any of the above-described methods.
[0042] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects:
[0043] An active learning process is added, that is, the uncertainty measurement value of each second historical fault vibration signal is determined according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model. By determining the uncertainty measurement value, it is helpful to screen the second historical fault diagnosis signal that is not labeled with a fault label, thereby achieving effective expansion of the training samples and improving the training accuracy of the model. By screening out samples with high uncertainty, the model can focus more on learning those representative features that contribute more to classification or prediction tasks. This helps to enhance the generalization ability of the model and enable the model to better adapt to new data and unknown situations. In addition, by continuously iteratively updating the one-dimensional fault diagnosis model to continuously improve its performance, after obtaining the one-dimensional fault diagnosis model of each part, the fault category obtained by the one-dimensional fault diagnosis model of each part can be voted, and the fault category with the most votes is selected as the final fault diagnosis result. This method realizes the comprehensive judgment of multiple models. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:
[0045] Figure 1 A schematic diagram of a device fault diagnosis method based on active learning provided in an embodiment of this specification;
[0046] Figure 2A fault diagnosis model architecture diagram of a one-dimensional vibration signal convolutional neural network provided in an embodiment of this specification;
[0047] Figure 3 A schematic diagram of an active learning sampling process provided in an embodiment of this specification;
[0048] Figure 4 A schematic diagram of sample uncertainty calculation provided in an embodiment of this specification;
[0049] Figure 5 A schematic diagram of the structure of a device fault diagnosis device based on active learning provided in an embodiment of this specification;
[0050] Figure 6 A schematic diagram of the structure of a non-volatile storage medium provided in an embodiment of this specification. DETAILED DESCRIPTION
[0051] The embodiments of this specification provide a device fault diagnosis method, device and medium based on active learning.
[0052] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be described clearly and completely below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.
[0053] like Figure 1 As shown, the embodiment of this specification provides a flow chart of a device fault diagnosis method based on active learning. Figure 1 It can be seen that in one or more embodiments of this specification, a device fault diagnosis method based on active learning includes:
[0054] S101: Determine whether a one-dimensional fault diagnosis model trained based on a first historical fault diagnosis signal meets preset model performance; wherein the first historical fault vibration signal is a fault vibration signal marked with a fault label on a single part of the equipment to be detected.
[0055] The amount of data of the equipment in normal operation is much higher than the data of fault samples, while the amount of common fault sample data is greater than that of uncommon fault samples, resulting in a small effect of standard samples on improving model performance. In order to solve the problem of low accuracy of model fault diagnosis obtained with a small amount of labeled data, this manual first determines whether the one-dimensional fault diagnosis model trained with the first historical fault diagnosis signal meets the pre-set model performance requirements. Among them, it should be noted that the first historical fault vibration signal is a fault vibration signal marked with a fault label on a single part of the equipment to be detected. This process first determines whether the one-dimensional fault diagnosis model meets the preset performance requirements, and makes subsequent adjustments and optimizations accordingly. It is an effective strategy to solve the problem of low accuracy of model fault diagnosis with a small amount of labeled data, and is of great significance to improving the overall equipment management and maintenance level.
[0056] Further, in one or more embodiments of the present specification, before determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes:
[0057] like Figure 2 As shown, the first historical fault diagnosis signal is input into the one-dimensional convolution layer of the initial one-dimensional fault diagnosis model, so as to traverse the first historical fault vibration signal according to the fixed stride of the one-dimensional convolution layer and the preset convolution kernel, and obtain the convolution output vector corresponding to the first historical fault diagnosis signal. That is to say, the input is traversed with a fixed stride, and the value corresponding to the rolled area is multiplied and summed each time. Then the convolution moves according to the step size, and the operation is repeated until the convolution kernel traverses the entire input area. Assuming that the lth layer is a convolution layer, the one-dimensional convolution operation formula of this layer is:
[0058]
[0059] In the formula, is the vector obtained after the j-th convolution calculation of layer l; M is the number of input feature vectors; is the i-th input feature vector of layer l; Represents related operations; The jth convolution kernel of the lth layer convolves with the i-th input feature vector; is the j-th bias vector of layer l.
[0060] Then as Figure 2As shown in the figure, the one-dimensional output vector x corresponding to the convolution output vector is determined according to the fully connected layer of the initial one-dimensional fault diagnosis model, and the one-dimensional output vector x is input into the softmax layer of the initial one-dimensional fault diagnosis model to determine the probability of each fault category corresponding to the one-dimensional output vector. That is to say, in the softmax layer, the output of the previous fully connected layer is a one-dimensional vector x. Assuming that the fault category label is y∈{1,2,3…,K}, the probability that the sample x belongs to category k is:
[0061]
[0062] Where θ is all the training parameters in the softmax model, θ = [θ1, θ2, θ3…, θ K ]; is the normalization function.
[0063] Then, the output result of the initial one-dimensional fault diagnosis model is determined according to the probability of the fault category, and then it is determined whether to iteratively train the initial one-dimensional fault diagnosis model according to the output result to obtain the trained one-dimensional fault diagnosis model. In this process, the one-dimensional fault diagnosis model is first trained based on the first historical fault diagnosis signal with a fault label. During the training process, the accuracy and robustness of the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal are improved through the advantages of effective feature extraction, intuitive probability output, flexibility of iterative training, and targeted model optimization.
[0064] Further, in one or more embodiments of the present specification, after determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes:
[0065] like Figure 2 If it is determined that the trained one-dimensional fault diagnosis model meets the preset model performance, then the trained one-dimensional fault diagnosis model is used as the one-dimensional fault diagnosis model of the corresponding part. Then the fault diagnosis signal of the corresponding part of the equipment to be detected is input into the one-dimensional fault diagnosis model to obtain the fault diagnosis result of the fault diagnosis signal. By judging the model performance, the problem of wasting computing resources in the redundant iteration process is avoided.
[0066] S102: If not, then determine the uncertainty measurement value of each of the second historical fault vibration signals according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model; wherein the second historical fault diagnosis signal is a fault vibration signal on a single part of the equipment to be detected that is not marked with a fault label.
[0067] After obtaining the trained one-dimensional fault diagnosis model based on the above step S101, if it is determined that the trained one-dimensional fault diagnosis model does not meet the preset model performance, then at this time, in order to avoid the need for a large amount of labeled data as support for the training of the deep learning model. However, the occurrence of equipment failures is often sporadic, and the fault types are diverse and difficult to fully cover, resulting in the problem that it is difficult to train a high-quality fault diagnosis model in a deep learning method under limited data. In the embodiment of this specification, an active learning process is added, that is, the uncertainty measurement value of each second historical fault vibration signal is determined according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model. Among them, it should be noted that the second historical fault diagnosis signal is a fault vibration signal on a single part of the equipment to be detected that is not marked with a fault label. By determining the uncertainty measurement value, it is helpful to screen in the second historical fault diagnosis signal that is not marked with a fault label, thereby achieving effective expansion of the training sample and improving the training accuracy of the model.
[0068] Specifically, in one or more embodiments of the present specification, according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model, determining the uncertainty measurement value of each second historical fault vibration signal specifically includes the following process:
[0069] First, according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model, determine the probability and fault diagnosis result of each fault category corresponding to the second historical fault diagnosis signal. Then, according to the probability of each fault category, determine the maximum probability of the fault category of the trained one-dimensional fault diagnosis model at the location. Sum the difference between the preset data and the maximum probability corresponding to each location to obtain the initial first uncertainty measurement value, and normalize the initial first uncertainty measurement value to obtain the first uncertainty measurement value. Then, according to the fault diagnosis result, determine the model prediction probability corresponding to the trained one-dimensional fault diagnosis model, and perform standardized summation on the variance of the model prediction probability corresponding to each location to obtain the initial second uncertainty measurement value, and normalize the initial second uncertainty measurement value to obtain the second uncertainty measurement value. Then combine the first uncertainty measurement value and the second uncertainty measurement value to determine the uncertainty measurement value of each second historical fault vibration signal. That is, combine the following Figure 4 As shown, sampling is performed using uncertainty measurement. The following is the uncertainty measurement of the sample:
[0070]
[0071] In the formula, g x represents the uncertainty of sample x, K represents the number of fault category labels, and M represents the number of sampling locations, i.e. the number of training models. represents the maximum fault category probability obtained by the model sample x at the i-th location, It means that at the jth part of sample x, the model obtains the variance of the probability of each fault category for the sample, It represents the maximum variance of the probability of each fault category obtained by the model for all samples at the jth part. In the above formula, it mainly includes two parts. The first part calculates the sum of 1 minus the maximum category probability for each part, and normalizes it as the uncertainty of the first part. The larger the value, the higher the uncertainty of the sample; the second part calculates the normalized sum of the variance of the model prediction probability of each part, and normalizes it as the uncertainty of the second part. The larger the value, the higher the uncertainty of the sample. Therefore, g x It is between 0 and 1, and the larger the value, the higher the uncertainty of the sample.
[0072] S103: screening the second historical fault diagnosis signals according to the uncertainty measurement value of each of the second historical fault diagnosis signals, so as to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signals.
[0073] According to the uncertainty measurement value of each second historical fault diagnosis signal obtained in step S102, downsampling of unlabeled samples is achieved, that is, Figure 3 As shown, the second historical fault diagnosis signal is screened to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signal. By screening the second historical fault diagnosis signal by the uncertainty metric, samples with high uncertainty and small contribution to model training can be removed. In addition, in machine learning and deep learning, the size of the training data directly affects the training time and computational cost of the model. By downsampling, that is, reducing the amount of training data, the computational cost can be significantly reduced. Since the uncertainty metric helps us identify samples that contribute less to model training, the amount of training data can be reduced and the training efficiency can be improved while ensuring the performance of the model. And by screening out samples with high uncertainty, the model can focus more on learning those representative features that contribute more to classification or prediction tasks. This helps to enhance the generalization ability of the model and enable the model to better adapt to new data and unknown situations.
[0074] Specifically, in one or more embodiments of the present specification, according to the uncertainty measurement value of each second historical fault diagnosis signal, the second historical fault diagnosis signal is screened to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signal, specifically including:
[0075] According to the uncertainty measurement value of each second historical fault diagnosis signal, the second historical fault diagnosis signal is sorted. Then, the second historical fault diagnosis signal of the preset ratio is obtained in sequence according to the sorting order. The second historical fault diagnosis signal of the preset ratio is sent to the preset expert end for labeling to obtain the second historical fault diagnosis signal marked with the fault label, and the second historical fault diagnosis signal of the preset ratio is updated with the first historical fault diagnosis signal. In the field of fault diagnosis, the data of normal operating status is often far more than the fault sample data, and there is a distinction between common and uncommon fault samples. This data imbalance problem will affect the training effect of the model. Downsampling the unlabeled samples by the uncertainty measurement value can alleviate the data imbalance problem to a certain extent, make the model pay more attention to those fault types that are difficult to distinguish, and improve the diagnostic accuracy of the model. For the preset ratio, the samples with the top 2% of the uncertainty value can be selected for labeling, and the labeled samples can be added to the training set, the model can be retrained and updated; at the same time, the unselected samples can be updated with the unlabeled sample set.
[0076] S104: Based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain a one-dimensional fault diagnosis model for each part, so as to identify and diagnose the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part.
[0077] According to the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain the one-dimensional fault diagnosis model of each part, so as to identify and diagnose the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part. This process not only focuses on the fault diagnosis of a single part, but also obtains the one-dimensional fault diagnosis model of each part through iterative updating. This enables the model to support the fault diagnosis results of multiple parts, and the diagnosis results of the equipment to be tested are determined to improve the comprehensiveness and accuracy of fault diagnosis.
[0078] Specifically, in one or more embodiments of the present specification, based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain the one-dimensional fault diagnosis model of each part, specifically including:
[0079] First, the trained one-dimensional fault diagnosis model is trained according to the updated first historical fault diagnosis signal, and the second historical fault diagnosis signal corresponding to the fault vibration signal not labeled with the fault label is updated according to the screened second historical fault diagnosis information. Then, the training set is iteratively updated, so that the trained one-dimensional fault diagnosis model is iteratively updated according to the updated training set to obtain the one-dimensional fault diagnosis model of each part that meets the requirements. In this process, by continuously iteratively updating the training set, the model can gradually adapt to more extensive and complex fault conditions. The updated first historical fault diagnosis signal contains more representative fault features, and the screened second historical fault diagnosis signal supplements the unlabeled but potentially valuable samples, which are helpful to improve the model's ability to identify various types of faults. During the iterative update process, the model will continue to be exposed to new and diverse data, which helps to reduce the overfitting phenomenon of the model to a specific data set. At the same time, by screening samples with lower uncertainty metrics, the impact of noise and abnormal data on model training can be further reduced, thereby enhancing the robustness of the model. In addition, a one-dimensional fault diagnosis model of each part is obtained through iterative updating, so that the model can perform independent fault diagnosis on different parts of the equipment, and can also summarize the judgment results of each part to obtain the equipment fault diagnosis results.
[0080] Furthermore, in one or more embodiments of the present specification, before identifying and diagnosing the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part, the method further includes:
[0081] Based on the preset acquisition devices of each part of the equipment to be detected, the fault vibration signals of each part are acquired based on the time sequence. Then, the fault vibration signals are segmented based on the input size corresponding to the one-dimensional fault diagnosis model of each part to obtain the time series data of the vibration signal to be input. Furthermore, in one or more embodiments of the present specification, the fault vibration signals of each part are identified and diagnosed based on the one-dimensional fault diagnosis model of each part, specifically including:
[0082] The time series data of the vibration signal to be input is input into the one-dimensional fault diagnosis model of the corresponding part, so that the fault category of the fault vibration signal is determined according to the probability of each fault category output by the one-dimensional fault diagnosis model of the corresponding part. The fault categories obtained by the one-dimensional fault diagnosis model of each part are voted, and the fault category with the most votes is selected as the final fault diagnosis result. For a large number of fault vibration signals collected, by segmenting and processing into time series data that meets the model input size, the one-dimensional fault diagnosis model can be efficiently used for processing. This processing method not only reduces the consumption of computing resources, but also improves the speed and efficiency of diagnosis. By voting on the fault categories obtained by the one-dimensional fault diagnosis model of each part, the fault category with the most votes is selected as the final fault diagnosis result. This method realizes the comprehensive judgment of multiple models. Since the models of different parts may capture fault characteristics from different angles and levels, the comprehensive judgment can further enhance the reliability and accuracy of the diagnosis results.
[0083] like Figure 5 As shown, the embodiment of this specification provides a structural schematic diagram of a device fault diagnosis device based on active learning. Figure 5 It can be seen that in one or more embodiments of this specification, a device fault diagnosis device based on active learning includes:
[0084] at least one processor; and,
[0085] a memory communicatively connected to the at least one processor; wherein,
[0086] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above methods
[0087] like Figure 6 As shown, the present specification provides a schematic diagram of the structure of a non-volatile storage medium. Figure 6 It can be seen that in one or more embodiments of the present specification, a non-volatile storage medium stores computer executable instructions, and the computer executable instructions can execute any of the above-described methods.
[0088] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part description of the method embodiment.
[0089] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0090] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.
Claims
1. A device fault diagnosis method based on active learning, characterized in that: The method comprises: Determine whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance; wherein the first historical fault vibration signal is a fault vibration signal marked with a fault label on a single part of the equipment to be detected; If not, then determining the uncertainty measurement value of each of the second historical fault vibration signals according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model; wherein the second historical fault diagnosis signal is a fault vibration signal on a single part of the equipment to be detected that is not marked with a fault label; screening the second historical fault diagnosis signals according to the uncertainty measurement value of each of the second historical fault diagnosis signals, so as to update the first historical fault diagnosis signal according to the screened second historical fault diagnosis signals; Based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain a one-dimensional fault diagnosis model for each part, so as to identify and diagnose the fault vibration signal of each part based on the one-dimensional fault diagnosis model of each part.
2. The device fault diagnosis method with active learning according to claim 1 is characterized in that: Determining the uncertainty measurement value of each of the second historical fault vibration signals according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model specifically includes: Determining the probability and fault diagnosis result of each fault category corresponding to the second historical fault diagnosis signal according to the output result of the second historical fault vibration signal output by the trained one-dimensional fault diagnosis model; Determine the maximum probability of the fault category of the trained one-dimensional fault diagnosis model at the location according to the probability of each fault category; Summing the difference between the preset data and the maximum probability corresponding to each part to obtain an initial first uncertainty measurement value, and normalizing the initial first uncertainty measurement value to obtain a first uncertainty measurement value; Determining the model prediction probability corresponding to the trained one-dimensional fault diagnosis model based on the fault diagnosis result, performing standardized summation on the variances of the model prediction probabilities corresponding to the various parts to obtain an initial second uncertainty measurement value, and performing normalization processing on the initial second uncertainty measurement value to obtain a second uncertainty measurement value; The uncertainty metric value of each of the second historical fault vibration signals is determined by combining the first uncertainty metric value and the second uncertainty metric value.
3. The device fault diagnosis method based on active learning according to claim 1 is characterized in that: Before determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes: Inputting the first historical fault diagnosis signal into the one-dimensional convolution layer of the initial one-dimensional fault diagnosis model, traversing the first historical fault vibration signal according to the fixed stride of the one-dimensional convolution layer and the preset convolution kernel, and obtaining a convolution output vector corresponding to the first historical fault diagnosis signal; Determine the one-dimensional output vector corresponding to the convolution output vector based on the fully connected layer of the initial one-dimensional fault diagnosis model, and input the one-dimensional output vector into the softmax layer of the initial one-dimensional fault diagnosis model to determine the probability of each fault category corresponding to the one-dimensional output vector; The output result of the initial one-dimensional fault diagnosis model is determined based on the probability of each of the fault categories, and it is determined whether to perform iterative training on the initial one-dimensional fault diagnosis model according to the output result to obtain a trained one-dimensional fault diagnosis model.
4. The device fault diagnosis method based on active learning according to claim 1 is characterized in that: After determining whether the one-dimensional fault diagnosis model trained based on the first historical fault diagnosis signal meets the preset model performance, the method further includes: If it is determined that the trained one-dimensional fault diagnosis model meets the preset model performance, the trained one-dimensional fault diagnosis model is used as the one-dimensional fault diagnosis model of the corresponding part; The fault diagnosis signal of the corresponding part of the equipment to be detected is input into the one-dimensional fault diagnosis model to obtain the fault diagnosis result of the fault diagnosis signal.
5. The device fault diagnosis method based on active learning according to claim 1 is characterized in that: The second historical fault diagnosis signals are screened according to the uncertainty measurement values of the second historical fault diagnosis signals to update the first historical fault diagnosis signals according to the screened second historical fault diagnosis signals, specifically comprising: sorting the second historical fault diagnosis signals based on the uncertainty measure value of each of the second historical fault diagnosis signals; sequentially obtaining a preset proportion of second historical fault diagnosis signals based on the sorting order; The preset proportion of the second historical fault diagnosis signal is sent to a preset expert terminal for labeling to obtain a second historical fault diagnosis signal labeled with a fault label, and the preset proportion of the second historical fault diagnosis signal is used to update the first historical fault diagnosis signal.
6. The device fault diagnosis method based on active learning according to claim 1 is characterized in that: Based on the updated first historical fault diagnosis signal and the screened second historical fault diagnosis signal, the trained one-dimensional fault diagnosis model is iteratively updated to obtain a one-dimensional fault diagnosis model for each part, specifically including: The trained one-dimensional fault diagnosis model is trained based on the updated first historical fault diagnosis signal, and the second historical fault diagnosis signal corresponding to the fault vibration signal not marked with a fault label is updated according to the filtered second historical fault diagnosis information; The training set is iteratively updated to iteratively update the trained one-dimensional fault diagnosis model based on the updated training set, so as to obtain a one-dimensional fault diagnosis model of each part that meets the requirements.
7. The device fault diagnosis method based on active learning according to claim 1 is characterized in that: Before identifying and diagnosing the fault vibration signals of each part based on the one-dimensional fault diagnosis model of each part, the method further includes: Based on the preset collection devices of various parts of the equipment to be detected, the fault vibration signals of various parts are obtained by collecting data based on the time sequence; The fault vibration signal is segmented based on the input size corresponding to the one-dimensional fault diagnosis model of each part to obtain the time series data of the vibration signal to be input.
8. The device fault diagnosis method based on active learning according to claim 7 is characterized in that: Based on the one-dimensional fault diagnosis model of each part, the fault vibration signal of each part is identified and diagnosed, including: Inputting the time series data of the vibration signal to be input into the one-dimensional fault diagnosis model of the corresponding part, so as to determine the fault category of the fault vibration signal according to the probability of each fault category output by the one-dimensional fault diagnosis model of the corresponding part; The fault categories obtained by the one-dimensional fault diagnosis model of each part are voted, and the fault category with the most votes is selected as the final fault diagnosis result.
9. A device fault diagnosis device based on active learning, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: execute any of the methods described in claims 1-8.
10. A non-volatile storage medium storing computer executable instructions, characterized in that: The computer executable instructions can: execute the method described in any one of claims 1 to 8.